Comparison and improvements of optimization methods for gas emission source identification

Abstract

Identification of gas leakage source term is important for atmosphere safety. Optimization is one useful method to determine leakage source parameters. The performances of different optimization methods, including genetic algorithm (GA), simulated annealing (SA), pattern search (PS) method, Nelder–Mead simplex method (N–M simplex) and their hybrid optimization methods, were discussed. It was seen that GA–PS hybrid optimization has the best performance for location and source strength estimation while the hybrid methods with N–M simplex is the best one when time cost and robustness are added into consideration. Moreover, the performances of these optimization methods with different initial values, signal noise ratios (SNR), sensor numbers and sensor distribution forms were discussed. Further, experiment data test showed that the less deviation of forward simulation model from the real condition, the better performance of the source parameters determination method is. When two error correction coefficients were added to the Gaussian dispersion model, the accuracy of source strength and downwind distance estimation is increased. Other different cost functions were also applied to identify the source parameters. Finally, a new forward dispersion model based on radial basis function neural network and Gaussian model (Gaussian–RBF network) was presented and then it was applied to determine the leakage source parameters. The results showed that the performance of optimization method based on Gaussian–RBF network model is significantly improved, especially for location estimation. Therefore, the optimization method with a good selection of forward dispersion model and cost function will obtain a satisfactory estimation result.